Table 4.
The model performance of different cut-off thresholds in the test dataset.
| Algorithms | Cut-Off Threshold | Sensitivity | Specificity | Precision | True Positive |
True Negative |
False Positive |
False Negative |
|---|---|---|---|---|---|---|---|---|
| RF | 0.3 | 82.9% | 69.2% | 51.6% | 237 (23.6%) | 498 (49.5%) | 222 (22.1%) | 49 (4.9%) |
| 0.4 | 69.9% | 85.8% | 66.2% | 200 (19.9%) | 618 (61.4%) | 102 (10.1%) | 86 (8.6%) | |
| 0.41 | 68.2% | 86.0% | 65.9% | 195 (19.4%) | 619 (61.5%) | 101 (10.0%) | 91 (9.1%) | |
| 0.5 | 57.7% | 94.0% | 79.3% | 165 (16.4%) | 677 (67.3%) | 43 (4.3%) | 121 (12.0%) | |
| 0.53 | 51.0% | 96.5% | 85.4% | 146 (14.5%) | 695 (69.1%) | 25 (2.5%) | 140 (13.9%) | |
| 0.6 | 38.8% | 98.2% | 89.5% | 111 (11.0%) | 707 (70.3%) | 13 (1.3%) | 175 (17.4%) | |
| 0.7 | 21.3% | 99.6% | 95.3% | 61 (6.1%) | 717 (71.3%) | 3 (0.3%) | 225 (22.3%) | |
| XGBoost | 0.3 | 89.9% | 48.8% | 41.1% | 257 (25.5%) | 351 (34.9%) | 369 (36.7%) | 29 (2.9%) |
| 0.4 | 82.2% | 64.3% | 47.8% | 235 (23.4%) | 463 (46.0%) | 257 (25.5%) | 51 (5.1%) | |
| 0.41 | 80.8% | 67.0% | 49.4% | 231 (23.0%) | 483 (48.0%) | 237 (23.6%) | 55 (5.4%) | |
| 0.5 | 70.6% | 77.5% | 55.5% | 202 (20.1%) | 558 (55.5%) | 162 (16.1%) | 84 (8.3%) | |
| 0.53 | 67.1% | 81.0% | 58.4% | 192 (19.1%) | 583 (58.0%) | 137 (13.6%) | 94 (9.3%) | |
| 0.6 | 53.5% | 90.7% | 69.5% | 153 (15.2%) | 653 (64.9%) | 67 (6.7%) | 133 (13.2%) | |
| 0.7 | 33.2% | 97.9% | 86.3% | 60 (5.9%) | 693 (68.9%) | 5 (0.5%) | 248 (24.7%) |
XGBoost: eXtreme Gradient Boosting, RF: random forest.